/SaTNC-A-Transformer-baed-neural-network-for-creep-rupture-life-prediction

Primary LanguagePython

SaTNC: A Transformer-baed neural network for creep rupture life prediction

By Fan Yang, Wenyue Zhao.

Background

With the development of material informatics, it becomes more important to introduce physical information constraints into the ML model. Here, a neural network with physical information constraints (Transformer-based) is designed for the prediction of creep rupture life, which is concerned in alloy design.

Updata

05/2/2023 Initial commits:

  1. Creep data, including creep datasets (.csv).
    Note, except alloying elements features (wt.%), the features with "_L12" and "_A1" postfix are contents (at.) of alloying elements in γ'/γ, which calculated by ThermoCalc.
  2. SaTNC model code
  3. ML model, including SVR, RF, LightGBM, DCSA (refer to https://github.com/wujunming1/mla-shu)
  4. The processing data: Elemental representation in the stage of Feature Fusion and processing code is provided in "Processing" file

Usage

The versions of the pyhton library used are as follows:
pandas -- 1.3.1
numpy -- 1.20.3
scikit-learn -- 1.1.2
lightgbm -- 3.2.1
torch -- 1.9.0